AI transformation of an automotive group's office

The situation

A large automotive group with an office of several hundred staff came to us. Leadership's question was direct: "How do we adopt AI so it doesn't turn into yet another dead pilot?" The market backdrop is well known — according to MIT, 95% of organizations see no return on their generative-AI pilots. The cause is rarely the technology or the budget; it is an adoption model set up the wrong way.

The classic failure we wanted to rule out: a wave from the top, rollout from the bottom, and leaders who never use AI themselves — so everyone else sees an instruction, not a direction. The group had already hit this wall in earlier digitalization waves.

Diagnostics and approach

Rather than scaling the tools straight away, we proposed running a first phase — a managed entry into the transformation. The logic: first build leaders' personal practice and a baseline AI culture, then select use cases and test their value, and only then move to industrial rollout and process redesign.

The anchor model is ADKAR: awareness, desire, knowledge, ability, reinforcement. This is not textbook theory — it is an operating framework that keeps change from fizzling out in the organization's middle layer.

The solution

Top down: leaders use AI personally

The CEO and vice presidents receive one-on-one support on their real work: preparing for committee meetings, reviewing reporting, working with documents. The goal is not to "train them on AI" but to build a habit — for a given task, open the AI assistant first, then act.

Assistants for executive aides and key specialists

Executive assistants are a natural entry point: they already manage the information flows of the senior leadership. From there it extends to legal, finance, the commercial unit, and HR. Each function has its own use cases and its own quality bar.

A map of potential use cases

In parallel with the training, we build a diagnostic funnel across 8 office functions: legal work (contract review, regulatory monitoring), sales and the dealer network, people and learning, finance and controlling, procurement, marketing, executive support, and IT. For each, we capture the real pain points, the owners of the impact, and the data and information-security constraints.

One example — reviewing supplier contracts

Before: a lawyer reads the contract by hand, checks it against the internal playbook from memory, notes the deviations, and consults a partner on any non-standard wording. Typical turnaround — one to two business days per contract. Partner time is spent even on the routine items.

After: the contract is loaded into an AI assistant that checks it against the company playbook and returns a list of deviations and suggested edits. The lawyer makes the calls, refines the draft, and issues the final version. Typical turnaround drops from one or two days to a few hours. The partner steps in only for the non-standard. A consistent position is now held by the playbook, not by individual experts.

Safe rules and the Russian IT perimeter

A separate workstream covers the rules for using AI under Federal Law 152-FZ, information-security requirements, and corporate compliance. Which data is permitted in which environments, which models may be used, and where anonymization is mandatory. Without that frame, any personal practice quickly runs into a stop signal from the lawyers.

What the first phase covered

• Hands-on support for senior leaders and their teams on real tasks
• Assistants for executive aides and key senior specialists
• Change-management discipline: sponsorship, evangelists at every level, reinforcement of practices
• A map of value sources and use cases — a guide for prioritization, not a promise of impact
• Diagnostic interviews with leadership — surfacing real pain points, constraints, and impact owners
• Rules for working with AI inside the Russian perimeter: 152-FZ, information security, compliance, permitted models and environments

What stayed out of the first phase

We deliberately left out of phase one: industrial rollout of AI agents and autonomous processes, custom integrations with 1C/CRM, capturing the production perimeter, and building proprietary models. These are later phases — once leadership has mastered the tools and articulated where the value lies.

Results of the first phase

The group walks away with: AI-readiness across leadership (able to frame tasks, judge quality, see the limits), a map of prioritized use cases, an AI adoption program for four audiences (leaders, business sponsors, assistants, key specialists), and a change model that lowers the risk of getting stuck in perpetual pilot mode.

On that foundation, phase two begins — selecting and piloting use cases through three lenses: desirability, technical feasibility, and economic viability. Complete with a brief, baseline metrics, an impact owner, and a build-or-buy decision.